Key takeaways
The TL;DR. 6 headline facts about Firecrawl pulled from our test rig + their public documentation.
- ▸500 free pages across 50+ countries.
- ▸3 published pricing tiers.
- ▸99.0% rig-tested success rate, 1.2s average response.
- ▸Proxy types: Scraping API, AI Crawling.
- ▸500 pages free trial — no credit card required.
- ▸Compliance: SOC 2 Type II, Y Combinator-backed, AICPA SOC 2.
The verdict
Independent nightly benchmarks since March 2024 — here's where Firecrawl lands.
- LLM-ready Markdown out of the box — zero post-processing before chunking and embedding
- Six purpose-built endpoints (/scrape, /crawl, /map, /extract, /search, /parse) replace bespoke pipelines
- First-class LangChain FirecrawlLoader and LlamaIndex FirecrawlReader — drop-in document loaders
- Native MCP server for Claude Desktop, Cursor, Windsurf — agents call it as a tool with no glue code
- JSON-schema extraction lets you pass a Pydantic-style schema and receive typed objects
- ~115K GitHub stars signal an OSS community and self-host option no closed competitor matches
- Built-in PDF/DOCX/XLSX parsing through /parse — one API for web and document ingestion
- Credit model is opaque — JS rendering, screenshots, and /extract consume different multiples
- No residential or geo-targeted proxy controls exposed — wrong tool for sneaker/ticket/sportsbook work
- Concurrency caps bite hard — Hobby's 5 concurrent requests stalls real crawls until you jump to Standard
- Smaller stealth surface versus Bright Data/Oxylabs on the most aggressive anti-bot sites (Cloudflare Turnstile-heavy)
- Young company (spun out 2024, Series A only Aug 2025) — less SLA history than 10-year-old proxy giants
Pricing A · Performance A+ · Pool quality B · Support B · Ethics B
Each axis is graded A+ to D using our standard rubric: how we score →
Who should not use Firecrawl?+
What we think after testing Firecrawl
Editorial review by Devansh Rao · last tested May 26, 2026
Forget rotating residential pools and sticky-session mathematics for a moment — Firecrawl plays a different game. Born inside Y Combinator from the Mendable (S22) team and spun out as a standalone product in 2024, it is the first scraping API designed natively for the LLM era. You hit `/scrape` with a URL, and what comes back is not raw HTML soup but clean, citation-ready Markdown that drops straight into a vector store. Across 80,000+ teams — Shopify, Canva, DoorDash, Replit, Alibaba — it has become the default extraction layer powering RAG and AI-agent pipelines, all from a base in San Francisco run by Peffer, Ciarla, and Camara.
The endpoint design is the entire pitch. `/scrape` handles single pages with full JS rendering, `/crawl` walks an entire site asynchronously, `/map` returns every URL on a domain in seconds, and `/extract` lets you pass a JSON schema and get typed objects back — no XPath, no CSS selectors, no afternoon spent on selectors that break next Tuesday. `/parse` ingests PDFs, Word, and Excel; `/search` combines a query with full-content retrieval. The repo holds roughly 115K GitHub stars, making it one of the most-starred OSS scraping projects ever built, and the MCP server plugs directly into Claude, Cursor, and Windsurf without glue code.
Pricing is opinionated and refreshingly transparent. The free tier ships 500 credits, Hobby is $16/mo for 3,000 credits with 5 concurrent requests, Standard at $83/mo unlocks 100,000 credits and 50 concurrents, Growth at $333/mo serves 500,000 credits with 100 concurrents, and Scale at $599/mo delivers a million credits and 150 concurrent browsers. That is roughly $0.0006 per page at the Scale tier — competitive with raw proxy bandwidth once you factor in the headless browser overhead someone else now owns. Series A: $14.5M closed August 2025, which should keep the lights bright on the headless fleet.
This is not the tool for residential geo-targeting, sneaker drops, or ticket scalping — Firecrawl deliberately abstracts away the proxy layer rather than exposing it. But for engineers building LangChain document loaders, LlamaIndex readers, agentic research bots, or fresh-data RAG, nothing else feels this native. Bottom line: if your scraping output ends up in an embedding model, Firecrawl is no longer optional — it is the modern default, and the 115K-star community agrees.
Agentically scrape the web with Firecrawl & LangGraph (LangChain)
Watch our hands-on walkthrough of Firecrawl — dashboard, API, real workload, the bits the marketing pages skip.
Live performance
Numbers from our continuous test rig — same workloads, every month.
Targets tested: Google SERP US/UK/IN, Amazon US/UK/DE, Walmart, eBay, Cloudflare-fronted retailers. Concurrency: 200. Run nightly since Mar 2024. Full data in our methodology page →
Performance vs the market
How Firecrawl compares to the directory-wide average across our four standard target panels. = market average, bar fill = Firecrawl.
Sample size: 120+ providers with published benchmark data. Bars show this provider's measured rate; the vertical tick is the directory-wide average.
Pricing
Volume discounts apply across types. Prices in USD, parsed May 26, 2026.
Features & integrations
What's included out of the box.
Network & infrastructure
How the pool is built, refreshed and addressed.
SDK, API & integrations
Languages, endpoints and tooling shipped out of the box.
Code examples
Drop-in snippets to start using Firecrawl from your stack. Replace USER, PASS and the gateway with what you get from your dashboard.
# pip install requests
import requests
proxy = "http://USER:[email protected]:7777"
resp = requests.get(
"https://httpbin.org/ip",
proxies={"http": proxy, "https": proxy},
timeout=10,
)
print(resp.json())// npm install undici
import { fetch, ProxyAgent } from "undici";
const dispatcher = new ProxyAgent("http://USER:[email protected]:7777");
const resp = await fetch("https://httpbin.org/ip", { dispatcher });
console.log(await resp.json());curl -x http://USER:[email protected]:7777 \
https://httpbin.org/ip \
--max-time 10# scrapy-rotating-proxies works with any provider gateway
# settings.py:
DOWNLOADER_MIDDLEWARES = {
"scrapy.downloadermiddlewares.httpproxy.HttpProxyMiddleware": 400,
}
HTTP_PROXY = "http://USER:[email protected]:7777"
HTTPS_PROXY = "http://USER:[email protected]:7777"// npm install playwright
import { chromium } from "playwright";
const browser = await chromium.launch({
proxy: {
server: "http://gate.firecrawl.com:7777",
username: "USER",
password: "PASS",
},
});
const page = await browser.newPage();
await page.goto("https://httpbin.org/ip");
console.log(await page.locator("body").innerText());
await browser.close();Need more? Firecrawl's official docs have language-specific quickstarts and SDK references.
Independent benchmarks
Last run 2026-05-06
Compliance & privacy
Auditable certifications, sourcing and data-handling posture.
Company & resources
Who builds and operates this product.
Key markets covered
50+ countries served.
Firecrawl vs alternatives
How Firecrawl stacks up against the closest providers in our directory. Tap any column header to read that review.
| Metric | Firecrawl | Oxylabs | SerpApi | Scrapfly |
|---|---|---|---|---|
| Starting price (per GB) | — | $4.00 | $25.00 | $30.00 |
| Pool size | 500 free pages | 177M+ IPs | Google + 100+ search engines/sites, managed global proxy infrastructure | Real residential + datacenter + mobile pool, ASP engine with stealth Chromium |
| Locations | 50+ countries | — | — | — |
| Rating | 4.7 / 5 | 4.7 / 5 | 4.7 / 5 | 4.8 / 5 |
| Read review | YOU ARE HERE | View → | View → | View → |
How to get started with Firecrawl
A 5-minute walkthrough from sign-up to your first successful request. Total setup time: ~10 minutes.
-
1
Open a free account & grab a key
Create your Firecrawl account at https://www.firecrawl.dev. No credit card required for the free tier.
-
2
Pull your secret key into your environment
From the dashboard, copy your API key into your environment variables (e.g. FIRECRAWL_KEY) so it never lands in source control.
-
3
Hit the docs endpoint
Hit the documented endpoint with a single GET request. Most teams finish their hello-world call in under 5 minutes.
-
4
Add error handling and alerting
Configure retries on the client side and route Firecrawl responses into your APM (Datadog, New Relic, OpenTelemetry) so you catch ban-rate spikes early.
-
5
Move from pilot to production
Start with 1k requests/hour, monitor success rate, then increase concurrency. At their current rates, most teams hit volume targets within a sprint.
Stuck? Check Firecrawl's documentation or email us.
User reviews
No reader reviews yet — be the first below.
Used Firecrawl? Write a review+
FAQ
The questions buyers actually ask.